Research

CURRENT PHD RESEARCH

Autonomous UGV navigation in unstructured terrain

My current PhD research focuses on enabling unmanned ground vehicles to navigate unknown and unstructured environments using onboard perception, terrain representations and real-time predictive control.

The central problem is not only to find a geometrically free route, but to reason about whether the terrain is traversable for a specific vehicle while the available map is incomplete and constantly changing.

These topics define my current doctoral research, while my broader interests extend across robotics, autonomous systems, perception, planning and control.

Research vision

My PhD connects the complete navigation chain rather than treating perception, planning and control as isolated components.

1 · Sense

RGB-D and LiDAR observations

2 · Represent

Local and global 2.5D terrain maps

3 · Reason

Traversability, risk and unknown space

4 · Navigate

Global guidance and local predictive control

The methods are designed around a clear principle: learned or data-driven components may provide information or guidance, while vehicle motion remains governed by explicit planning, control and safety mechanisms.

Published and accepted work

IROS 2026 · ACCEPTED

Observation-Conditioned Rollout Allocation for Sampling Model Predictive Control on Myopic Egocentric Elevation Maps

This work reallocates the sampling budget of a predictive controller according to the current egocentric terrain observation. It targets reactive navigation when the robot only sees a limited local map and must select useful motion hypotheses under real-time constraints.

Project page Publication record

Ongoing research directions

The following areas are active parts of my PhD. They are presented as ongoing research, not as completed publications.

Global terrain-aware navigation from aerial DEMs

Navigation using a global digital elevation model as prior information, combined with a live egocentric map acquired by the robot. The planner evaluates directional terrain properties, vehicle footprint, clearance, fall risk and unknown regions rather than treating the DEM as a conventional occupancy grid.

Current focus: global terrain guidance, local map updates and fast replanning when the prior map no longer matches the scene.

Reactive navigation on egocentric elevation maps

Local navigation with a limited sensor horizon and no assumption of a complete global map. The controller evaluates terrain geometry, collision risk, feasible corridors and progress while accounting for car-like vehicle constraints.

Current focus: terrain-aware MPPI, footprint validation, safer treatment of unknown space and robust obstacle avoidance.

Exploration and compact terrain memory

Exploration of bounded unknown areas using safe frontiers, terrain descriptors and compact snapshots of previously observed elevation maps. The objective is to retain useful terrain information without maintaining an unbounded dense map.

Current focus: frontier evaluation, compressed terrain memory and reuse of prior observations during exploration.

Goal-directed navigation without a prior map

A longer-term direction in which the robot receives a target rather than a predefined path and must combine egocentric perception, remembered terrain structure and explicit navigation mechanisms to reach it efficiently.

Research question: how much compact terrain memory is needed to make better local and global decisions without constructing a full dense map?

Experimental platform

The research is developed in simulation and transferred to a real car-like UGV equipped with onboard computing and multimodal perception.

Robot

Outdoor unmanned ground vehicle with car-like motion constraints

Perception

RGB-D cameras, 3D LiDAR, IMU and complementary sensing

Software

ROS 2 Humble, real-time C++ control and GPU-enabled onboard processing

Environment

Unknown, uneven and partially observable outdoor terrain

Evaluation priorities

Methods are assessed not only by whether the robot reaches the goal, but also by collision rate, minimum terrain margin, path length, execution time, smoothness, unknown-space exposure and real-time computational cost.

Interested in this work?

I am open to research discussions and collaborations in field robotics, autonomous systems, navigation, perception, planning, control and related robotics topics.

Email me LinkedIn